Challenges in Observing System Design: Perspectives from Antarctica to the North American Great Lakes
Date:
Seminar presented at the University of Washington on 15 October 2025 and at NOAA Pacific Marine Environmental Laboratory (PMEL) on 16 October 2025.
Overview
Environmental observations underpin our ability to understand and predict the Earth system. They inform weather forecasts, climate projections, ecosystem management, flood risk assessment, and water resource planning. Yet the ocean and Great Lakes are vast, dynamic systems, and the resources available for environmental monitoring are always limited.
This seminar examined the challenge of observing system design: how do we decide what to observe, where to observe it, and how to make the best use of limited observational resources?
Drawing on examples from the Southern Ocean and the North American Great Lakes, I reviewed a range of approaches used in observing system design, including Observing System Experiments (OSEs), Observing System Simulation Experiments (OSSEs), adjoint-based uncertainty quantification, and emerging machine learning methods. Particular attention was given to the role of covariance structures and connectivity in environmental systems, and how these relationships can be used to identify observations that provide the greatest information value.
The seminar concluded with recent work applying Convolutional Gaussian Neural Processes and the DeepSensor framework to Great Lakes observing system design, including the question: if funding is available for only one new observing platform, where should it be deployed?
Topics
- Observing system design
- OSEs and OSSEs
- Data assimilation
- Adjoint methods and uncertainty quantification
- Environmental covariance and connectivity
- Machine learning for sensor placement
- Convolutional Gaussian Neural Processes
- Great Lakes observing systems
- Southern Ocean observing systems
Take-Home Message
Observing system design is fundamentally a problem of making decisions under uncertainty. While modern tools such as data assimilation, adjoint methods, and machine learning can help evaluate observing strategies, practical constraints, scientific objectives, and stakeholder priorities remain essential parts of the design process.